HivemindOS manual

MiroShark And Runtime Gateways

MiroShark is the rehearsal room.

It is optional, but useful when a swarm plan needs simulation before real execution. HivemindOS also keeps a small runtime gateway layer for runtimes such as OpenClaw when they are used as agents in Fleet and Chat.

MiroShark

How it works:

Companion client: src/lib/services/miroshark/companion-client.ts.
Dashboard routes: /api/miroshark/status, /api/miroshark/manage, /api/miroshark/swarm, /api/miroshark/runs, /api/miroshark/analysis.
Swarm transformations live in src/features/swarm.

What MiroShark support can do:

Detect, install, start, and open MiroShark.
List templates and run metadata.
Start scenario simulations.
Accept /swarm-sim <scenario> launches from dashboard chat through the same /api/miroshark/swarm route used by the Swarm view.
Run hosted simulations from Swarm or native chat for $1.20 per run, paid from the user’s shared HivemindOS balance or directly from a selected crypto wallet through x402. The app remembers the payment choice while keeping the final price and confirmation visible before each run. If the run does not start, the user is not charged.
Keep local and self-hosted MiroShark available for free. Choosing a hosted run is the only action that creates the $1.20 cloud charge.
Fetch run status, summaries, reports, transcripts, exports, telemetry, market events, observability events, LLM calls, and intelligence where the companion exposes them.
Send simulation output back into agent or Kanban workflows.
Load archived runs from the shared vault.
Ask scenario-helper agents for template suggestions or scenario rewrites.
Analyze runs with a selected dashboard agent/runtime model.
Publish supported runs through the MiroShark experiment surface.

Swarm Theater

The Swarm view is the dashboard’s MiroShark workbench. It combines:

Template selection and template field validation.
Draft scenario, round count, and platform controls for Twitter/X, Reddit, Parallel, and Polymarket-style surfaces.
Live run loading states with progress labels and archive fallback.
Run timeline, observability, market, profile, action, and integration items transformed by src/features/swarm/swarm-transformers.ts.
Output panels for X threads, Reddit, Polymarket, research, ops, and integration views.
Analysis-agent selection backed by dashboard runtime/model selections.

The view is intentionally Hivemind-owned. MiroShark remains an optional companion service. HivemindOS owns the Fleet badge matching, route catalog fallback, archive routing, and shared-vault artifact handoff.

Dashboard chat also exposes /swarm-sim <scenario> for quick simulation launch. Use the Swarm view when the operator needs templates, fields, platform controls, analysis-agent selection, archive browsing, publishing, or deeper run inspection.

API Service Catalog

When MiroShark is discovered as a hivenet API service, /api/fleet/apps can attach a route catalog even if the service root is not interactive. The catalog prefers OpenAPI when available and otherwise falls back to the Hivemind-owned MiroShark signature. My Apps then shows grouped endpoints for templates, lifecycle, run data, graph, observability, and config.

Runtime Gateway Support

How it works:

Gateway client: src/lib/services/openclaw/gateway-client.ts.
Gateway health/token helper: src/lib/services/openclaw/gateway-health.ts.
The OpenClaw runtime adapter participates in the generic runtime adapter registry.
Chat is routed through the generic agent runtime path when an OpenClaw runtime profile is selected.

What runtime gateway support can do:

Runtime profile detection.
Runtime model selection from local config.
WebSocket chat bridge through the HivemindOS runtime chat surface.
Basic gateway token lookup for local runtime calls.